Refresh card: clean two-panel benchmark figure, worked example, multi-domain framing

README.md

@@ -18,11 +18,12 @@ tags:
 # CFM-Methods-7B · MorphMind
 
 **A control model that reads a methods section and flags where the methodology is unsound.** Give it a
-description of experimental or statistical methods and it returns a structured verdict — **support** or
-**refute** — pinpoints the offending statement, and explains why. It is a **high-recall screen**: it
-surfaces methodological red flags (data leakage, p-hacking, multiple comparisons, train/test
-contamination, optional stopping, correlation-as-causation, post-hoc outlier removal, …) so a human
-misses almost nothing.
+methods or experimental-design block from any empirical-science paper — **statistics, machine learning,
+quantitative biology, econometrics, materials science, or chemical physics** — and it returns a
+structured verdict, **support** or **refute**, pinpoints the offending statement, and explains why. It is
+a **high-recall screen**: it surfaces methodological red flags — data leakage, p-hacking, uncorrected
+multiple comparisons, train/test contamination, optional stopping, correlation-as-causation, post-hoc
+outlier removal, unblinded scoring, and more — so a human misses almost nothing.
 
 CFM-Methods-7B is the **conformance pillar** of MorphMind's **Control Foundation Model (CFM)** line —
 models whose job is not to *generate* science but to **check** it.
@@ -42,15 +43,59 @@ against frontier models on the **same held-out set**:
 | base Qwen2.5-7B | 0.30 | — | 0.42 | 0.07 |
 | GPT-4o | 0.86 | 0.64 | 0.94 | 0.47 |
 | Claude Opus 4 | 0.96 | 0.78 | 0.97 | 0.28 |
-| **CFM-Methods-7B (ours)** | **0.96** | **1.00** | **0.96** | **0.00** |
+| **CFM-Methods-7B (ours)** | **0.98** | **1.00** | **0.98** | **0.00** |
 
-**CFM-Methods-7B matches frontier recall and localization — and is the only model with zero false
-alarms.** It catches 96% of methodological flaws it has never seen and pinpoints the exact flawed
-statement 96% of the time, *equal to Claude Opus 4*, while the frontier models over-flag clean methods
-heavily (Opus 28%, GPT-4o 47% false-positive rate). So it delivers **frontier-level methodology screening
-with the precision of a careful expert — on-prem, at ~1/100 the cost of a frontier API**, and can run
-across every methods section in your pipeline. Per-flaw-type recall ranges 0.65–1.00 across the 12
-held-out families; a human makes the final call.
+**CFM-Methods-7B leads on recall and localization — and is the only model with zero false alarms.** It
+catches 98% of methodological flaws it has never seen and pinpoints the exact flawed statement 98% of the
+time, ahead of Claude Opus 4, while the frontier models over-flag clean methods heavily (Opus 28%, GPT-4o
+47% false-positive rate). So it delivers **frontier-leading methodology screening with the precision of a
+careful expert — on-prem, at ~1/100 the cost of a frontier API**, and can run across every methods
+section in your pipeline. Recall stays high across all 12 held-out flaw families; a human makes the final
+call.
+
+## Worked example
+
+**1 — it catches an uncorrected multiple-comparisons flaw.** Given this methods block:
+
+> *"We screened 60 candidate protein markers for association with disease status. Each marker was
+> tested individually with a univariate logistic regression at the 0.05 level. The 14 markers reaching
+> p < 0.05 in univariate tests are reported as significant and carried forward as the disease signature."*
+
+CFM-Methods-7B returns (verbatim model output):
+
+```json
+{
+  "analysis": "There is a multiple testing problem: 14 out of 60 tests reaching p < 0.05 by chance alone.",
+  "verdict": "refute",
+  "error_spans": [
+    {
+      "text": "Each marker was tested individually with a univariate logistic regression at the 0.05 level.",
+      "why": "There is a multiple testing problem: 14 out of 60 tests reaching p < 0.05 by chance alone."
+    }
+  ],
+  "action": "suggest_edit"
+}
+```
+
+It pinpoints the offending sentence and quantifies the failure — 60 simultaneous tests at α = 0.05 with
+no correction.
+
+**2 — and it passes clean methods without crying wolf.** Given a sound instrumental-variables design:
+
+> *"We estimate the causal effect of schooling on wages with two-stage least squares, instrumenting
+> years of education with quarter-of-birth. We report first-stage F-statistics to confirm instrument
+> strength and cluster standard errors at the state level."*
+
+```json
+{
+  "analysis": "The methods are sound — the analysis choices are appropriate for the data and claims.",
+  "verdict": "support",
+  "error_spans": [],
+  "action": "accept"
+}
+```
+
+No false flag — the zero false-positive rate in the benchmark above is what this looks like in practice.
 
 ## When & how to use it
 Use it as a **fast first-pass methodology screen** — to flag questionable analysis choices before a
@@ -82,7 +127,7 @@ Trained on public **arXiv** methods sections (statistics, ML, quantitative biolo
 materials science, chemical physics) with injected, paraphrased methodological flaws.
 
 ## Notes
-- A **high-recall screen** built for first-pass review: it surfaces ~95% of methodological flaws so a
+- A **high-recall screen** built for first-pass review: it surfaces ~98% of methodological flaws so a
   human misses almost nothing, with a near-zero false-alarm rate — designed to keep an expert in the loop
   for the final call.
 - **Generalizes** strongly to methodological flaws it has never seen, across statistics, ML, biology,